#source('/Volumes/MONOGAN/debtCeiling/data/debtMatch.R')

#clean up
rm(list=ls())

#set preferences
options(scipen=5,digits=6)

#set working directory (CHANGE THIS TO YOUR DIRECTORY)
setwd('/Volumes/MONOGAN/debtCeiling/data/')

#load libraries
library(foreign)
library(cem)
library(Matching)
library(MatchIt)
library(RItools)
library(MNP)
library(xtable)

#load data

full.data<-read.dta('debtCeilingCertified.dta')

#subsets
win.data<-subset(full.data,fullmatch==1)
share.data<-subset(full.data,sharematch==1)

### ANALYSIS WITH MATCHED SAMPLES TO TEST HYPOTHESES ###
#analysis of seat retention
win.rates<-table(win.data$no,win.data$winCert)
rownames(win.rates)<-c('yes vote','no vote')
colnames(win.rates)<-c('lost seat','kept seat')
win.rates
#table is for percentage of losers
#alternative hypothesis: greater number of losers among 'yes' voters than 'no' voters
prop.test(win.rates,alternative='greater')
#in fact, more 'no' voters did not retain their seat
#chi-squared (1 df)=1.3056, p=0.8734
#two-tailed p-value also shows no discernible effect either way

#analysis of two-party vote share
#alternative hypothesis: smaller share of the two-party vote among 'yes' voters than 'no' voters
t.test(incShare~no,data=share.data,alternative='less')
#difference: 64.7680-62.3615=2.4065
#t (108.352 df)=-1.5758, p=0.0590

### COMPARISON TO ANALYSIS WITH FULL DATA ###
#analysis of seat retention
win.rates.2<-table(full.data$no,full.data$winCert)
rownames(win.rates.2)<-c('yes vote','no vote')
colnames(win.rates.2)<-c('lost seat','kept seat')
win.rates
#table is for percentage of losers
#alternative hypothesis: greater number of losers among 'yes' voters than 'no' voters
prop.test(win.rates.2,alternative='greater')
#similar null result to matched analysis
#chi-squared (1 df)=0.4424, p=0.747

#analysis of two-party vote share
#alternative hypothesis: smaller share of the two-party vote among 'yes' voters than 'no' voters
t.test(incShare~no,data=full.data,alternative='less')
#difference: 67.4656-62.9790=4.4866
#t (201.386 df)=-3.7678, p=0.0001
#larger effect: almost surely overestimates the treatment effect

### ALTERNATE ESTIMATE OF TREATMENT EFFECT ###
library(MatchIt)
competitive<-subset(full.data,intermediate==1&!is.na(incShare)&!is.na(no),select=c(incShare,no,nominate1,obama08,inc_vote10,log_cash,inc_dem))
prop.out<-matchit(no~nominate1*obama08*inc_vote10*log_cash*inc_dem, data=competitive); summary(prop.out)
prop.data<-match.data(prop.out)
t.test(incShare~no,data=prop.data,alternative='less')

### GRAPHS OF DATA ###
pdf('matchedShare.pdf',width=4,height=4,pointsize=10)
plot(x=share.data$nominate1[share.data$no==1],y=share.data$incShare[share.data$no==1],xlim=c(-1,1),ylim=c(40,100),col='red',pch=2,xlab='NOMINATE (First Dimension)',ylab='Share of Two-Party Vote (2012)')
points(x=share.data$nominate1[share.data$no==0],y=share.data$incShare[share.data$no==0],xlim=c(-1,1),ylim=c(40,100),pch=20)
legend(x=0.5,y=100,c('yes','no'),pch=c(20,2),col=c('black','red'))
dev.off()

pdf('fullShare.pdf',width=4,height=4,pointsize=10)
plot(x=full.data$nominate1[full.data$no==1],y=full.data$incShare[full.data$no==1],xlim=c(-1,1),ylim=c(40,100),col='red',pch=2,xlab='NOMINATE (First Dimension)',ylab='Share of Two-Party Vote (2012)')
points(x=full.data$nominate1[full.data$no==0],y=full.data$incShare[full.data$no==0],xlim=c(-1,1),ylim=c(40,100),pch=20)
legend(x=0.5,y=100,c('yes','no'),pch=c(20,2),col=c('black','red'))
dev.off()

#percent retaining seat
win.rates
win.rates.2
pdf('winPercent.pdf',width=5,height=5,pointsize=10)
full.win<-matrix(c(83.2,80.4,86.1,78.4),byrow=FALSE,nrow=2,ncol=2); full.win
barplot(full.win,beside=TRUE,ylim=c(0,100), names.arg=c('All Members', 'Matched'), legend.text=c('yes','no'), ylab="Percent Retaining Seat", args.legend=list(pt.cex=.9,ncol=2))#x>2.5
abline(h=0,col='gray60')
box()
dev.off()

### MODEL OF INTERMEDIATE OUTCOMES ###
full.data.2<-subset(full.data,intermediate<6 & !is.na(abs_nom) & !is.na(no))
intermed.mnp<-mnp(intermediate~abs_nom+inc_vote10+obama08+log_cash+no,data=full.data.2); summary(intermed.mnp)
full.data.yes<-full.data.no<-full.data.2
full.data.yes$no<-0
full.data.no$no<-1
pred.yes<-predict(intermed.mnp,newdata=full.data.yes,type="prob") #voted yes
pred.no<-predict(intermed.mnp,newdata=full.data.no,type="prob") #voted no
probabilities<-rbind(apply(pred.yes$p,2,mean),apply(pred.no$p,2,mean))

pdf('intermedProb.pdf',width=4,height=4,pointsize=10)
barplot(probabilities, beside=TRUE, ylim=c(0,.7), names.arg=c('Contested','Uncontested','Lost Prim.','Retired','Higher Office'),legend.text=c('yes','no'),cex.names=.7,ylab="Predicted Probability")
abline(h=0,col='gray60')
box()
dev.off()
